Self-supervised Object Motion and Depth Estimation from Video

Abstract

We present a self-supervised learning framework to estimate the individual object motion and monocular depth from video. We model the object motion as a 6 degree-offreedom rigid-body transformation. The instance segmentation mask is leveraged to introduce the information of object. Compared with methods which predict pixel-wise optical flow map to model the motion, our approach significantly reduces the number of values to be estimated. Furthermore, our system eliminates the scale ambiguity of predictions, through employing the pre-computed camera egomotion and the left-right photometric consistency. Experiments on KITTI driving dataset demonstrate our system is capable to capture the object motion without external annotation, and contribute to the depth prediction in dynamic area. Our system outperforms earlier self-supervised approaches in terms of 3D scene flow prediction, and produces comparable results on optical flow estimation.

Towards Real-Time Monocular Depth Estimation for Robotics: A Survey

“…Thus, the network can be trained in an unsupervised manner through minimizing the photometric error. Dai et al [127] propose a self-supervised learning framework for jointly estimating individual object motion and depth from monocular video. Instead of modeling the motion by 2D optical flow or 3D scene flow, the object motion is modeled and predicted in the form of full 6 degrees of freedom (6 DoF).…”